Identification of membrane protein types via multivariate information fusion with Hilbert-Schmidt Independence Criterion

Membrane proteins perform a variety of functions vital to the survival of organisms, such as oxidoreductase, transferase or hydrolase. If the type of membrane protein can be detected, the function of protein can be quickly determined. Many existing computational methods not only use the autocorrelation function on the hydrophobicity index of amino acids, but also consider the evolutionary conservatism information of the primary protein sequences. In this study, we employ Average Blocks (AN/Block), Discrete Wavelet Transform (DWT), Discrete Cosine Transform (DCT), Histogram of Oriented Gradient (HOG) and Pseudo-PSSM (PsePSSM) to extract evolution characteristics from Position-Specific Score Matrix (PSSM). Then, we construct five kernels from above five corresponding feature sets. Finally, we propose a novel Multiple Kernel Support Vector Machine (MKSVM) classifier based on Hilbert Schmidt Independence Criterion (HSIC) to integrate five kernels for identifying membrane proteins. For the performance evaluation, our method is tested on four benchmark datasets of membrane proteins. The comparative results demonstrate that our prediction model achieves the best performance among all existing outstanding approaches. (C) 2019 Elsevier B.V. All rights reserved.